After having worked as an engineer for a couple of years, Ibe thought it was time for a change. Coincidence or not, a new opportunity presented itself that had an eye for both his passion for IT and education. He decided to take the plunge, joined the BeCode team in September, and is now preparing the first batch of Ghent AI students for their new career path. Curious about how Ibe integrates his background in robotics into the training or about the projects he is currently working on? Then be sure to read the interview below!
Can you tell me a bit more about yourself?
”I was working as an engineer at Duco, but started looking for a new challenge. My sister had contacts at BeCode and advised me to apply. In September I started working as an AI coach at BeCode Ghent. I was always interested in education, but I found the step too big. However, coaching those who want to start a new career in this field seemed like a step in the right direction. At BeCode they were not looking for someone with years of experience in education. The preference went rather to someone with the necessary professional knowledge who can also function well in a social environment. The position seemed to fit me perfectly, so I decided to apply. A brief interview with management soon followed and I was hired before I knew it.”
”Besides coaching juniors, I also work on several robotics projects. As these are time-consuming, I’m mainly focusing on small projects such as setting up (chat)bots. You don’t have to invest much time in these projects and you get a lot of entertainment in return.”
What robotics projects are you working on?
”Arai, a small robot that BeCode possesses, but that hasn’t been used a lot so far. In the first phase, I hope to be able to make a connection between her and our Discord server, which is more challenging and time-consuming than I first thought. The possibilities are endless with this robot and you don’t necessarily need a robotics background to program her. Some knowledge of robotics might be helpful, but it’s definitely not a requirement.”
How far along are you with programming the robot?
”I’ve managed to connect Arai to the Discord server, but due to insufficient time to train her, I’ve taken the connection offline for now. In the next phase, she will be programmed for marketing purposes. I hope to be able to connect the chatbot I developed to the robot, Arai, so she can conversate with people as well as answer their questions. In addition, I would also like to see her walking around the campus so she can interact with her surroundings.”
You indicated that you also work on small projects, such as developing bots. Where did the idea come from?
”In celebration of poetry week, each class group was asked to submit a poem. This seemed like a unique opportunity to send in partially AI-generated contributions. The initial goal of the bot was to generate poems for poetry week, but in the meantime, the bot has become a source of entertainment.”
What exactly do you have to do to generate a poem?
”Natural Language Processing relies on neural networks. A neural network is a fairly complex network of neurons that builds connections with other neurons in that same network. The ultimate goal of such a network, in the context of NLP, is to predict what the next word will be in a sequence of words. The network is trained using a dataset, which in this case comes from Wikipedia. By training the network, it will understand which words are related to each other and thus will be able to predict which word is next in the sequence. So if you were to enter the search term chess, the network will think of king, queen, rook, knight, etc. The poetry bot works similarly. The words that will appear will be poetry related.”
What stages did you go through in developing the bot?
”In addition to the poetry bot, I created a
chatbot. The chatbot and the poetry bot both rely on GPT-3, the latest NLP model from OpenAI. I just had to create an interface between our server and the API of GPT-3 so that you receive a response when you interact with the bot. The more input the bot receives, the more correct its prediction will be.”
”When making a prediction, the chatbot must take into account some character traits (sarcasm, friendly and helpful) and the previous conversations it has had with the user. But the larger the input sequence, the more expensive it is to generate output and the longer it takes for the user to receive a reply. I, therefore, set up the bot so that it only has to take into account the character traits as well as the last two interactions with the user when formulating an answer.”
So Python and knowledge of NLP are important when developing bots. What other technologies did you use in the development process?
”I actually used very few technologies. These bots have Discord as an interface and to set them up, you only need to be able to program in Python. Even knowledge of NLP is not really necessary, but it helps you understand what’s going on behind the scenes, how to increase the performance of your bot, and what context is important when making a prediction.”
”If you weren’t going to use GPT-3, there are a lot of other technologies that are also covered in the training that you can use. Logically you start with interpreting and processing your data. To build your model you can use tools like GPT-2. Designing and training your own network, in turn, can be done via Tensorflow or Pytorch.”
”During the last chapter of the training, the juniors are introduced to neural networks and how to train them using data. So they would be perfectly capable of realizing similar projects.”
Did the juniors help develop the bots?
”I designed the chatbot in one evening because the juniors had a week off and I had no one to talk to. The bot also replaced me during Simplon training – obviously in the proximity of a coach. The bot may be a fun tool, but it can never act as a full-fledged coach.”
”The chatbot, also called Ibot Denaux, is the little brother of the poetry bot. This bot does not generate poems. It is mainly intended for entertainment purposes, but it can also explain concepts. Nevertheless, it is important to take these answers with a grain of salt. The context of the bot is defined as “sarcastic,” so it lies and jokes quite easily. It’s fun for the students to see what’s possible with AI.”
What, in your opinion, is the added value for the juniors of such projects?
”This way, they can get acquainted with the technologies behind a chatbot. This kind of project should also show them that NLP can also be used for fun purposes. The goal of this project is actually twofold, to give them a target and to entertain them”
Can the other promotions also use your bots?
”The chatbot can only be used by the Ghent AI students, but it is possible to integrate it into the other servers. The poetry bot is openly accessible.”
Do you have plans to develop similar bots in the near future?
”I’m afraid that people would quickly get tired of me if I only develop bots. It doesn’t take much effort and it’s fun to make them, but I want to avoid overloading the servers. But who knows, if a nice idea pops up.”